Introduction of independent satellite bias correction on CO2 flux inversion

Takashi Maki¹, Keiichi Kondo², Kentaro Ishijima¹, Tsuyoshi, T. Sekiyama¹, Kazuhiro Tsuboi¹, and Takashi Nakamura²

1 : Meteorological Research Institute, Tsukuba, Japan

2 : Japan Meteorological Agency, Tokyo, Japan

TransCom 2024 meeting, 28^th May, 2024, Colorado (Hybrid)

Characteristics of in-situ CO2 observation

Conducted on the ground, ships, and aircraft, with very high accuracy (error of 0.1ppm or less)

Traceability is established under WMO/GAW　

　 World Calibration Center → District Calibration Centers → National Standards → 　　 Observatories, etc.

The number of stations is small compared to meteorological observations (less than 200 stations globally).

Generally, it takes a certain amount of time (several months to a year) to finalize the observed values.

Characteristics of satellite CO2 observation

Advantage

The same sensor can be used to observe a wide area.

A large number of observation data can be obtained.

Spatial representativeness is wide (1 to 10 km) and has high affinity with the model.

​

​

​

​

Disadvantage

Limitations of the observable area due to clouds, solar zenith angle, etc.　

Difficulty of nighttime observations except for thermal infrared (TIR)Retrieval.　

Biases adversely affect the results of data assimilation and inverse analysis.

There is a verification network (TCCON) using ground-based FTS, but it covers less than 30 sites globally (see above figure).

TCCON observation network (from NIES HP)

Analysis using satellite observation data

TCCON was used to perform a globally uniform correction (-1.2ppm), and an inverse analysis cross-comparison was performed (Takagi et al., 2014).

CO2 flux balance shifts with ground-based data inverse analysis (top left) and additional satellites (top right).

Analysis using satellite observation data

A CO2 balance analysis study using satellite observations (GOSAT, OCO-2, etc.) was conducted (Chevallier et al., 2019).

In many cases, the introduction of satellite observation data has significantly changed the CO2 balance compared to analyses based on field observations alone.

Bias of satellite observation data and its temporal and spatial variations are considered to be influential.

CO2 Inversion system

The entire globe is divided into 22 regions, and the CO2 flux in each region is determined by an evaluation function consisting of the difference between observed values (WDCGG) and a priori information to the smallest value.

The information has been provided once a year since 2009.

Plans to introduce satellite observation data (GOSAT) in the future.

https://ds.data.jma.go.jp/ghg/kanshi/info_kanshi.html

Satellite Bias correction system

Since it is difficult to evaluate and correct the bias of XCO2 using only surface observations (TCCON), we decided to evaluate and correct the bias of satellite observation data using the carbon dioxide distribution information provided by the independent inverse analysis.

We compared the difference between the two for 12 years (2009-2020) to evaluate the bias and trend.

A new inverse analysis was performed using the bias-corrected satellite observations and in-situ observations.

Feedback between observation and analysis (assimilation) is important.

Difference between satellite and independent inverse analysis

The interannual variation of the difference shows no significant trend, but seasonal variations are observed.

The average for the period was -1.26 ppm for global, -0.76 ppm for terrestrial, and -1.54 ppm for ocean.

The seasonal variations of the land and ocean appear to be inverted.

Seasonal change in difference (12-year average)

Seasonal and location-dependent differences were observed in the independent inversion.

The differences tended to be larger over land and at higher latitudes.

Satellite bias correction method

Five different bias correction methods were used to estimate the impact on the CO2 flux analysis.

Assume that XCO2_obs from satellite observations consists of the true value XCO2_true and the bias XCO2_bias where i and j are the east-west and north-south indices of the grid point, m is the month, and n is the year (assuming no trend in bias and random errors are negligible)

Averaged over N years, the result is as follows

Here, we cannot know the true value, but if N is large and is replaced by the value XCO2_inv after inverse analysis based only in-situ observations, the bias of the satellite observation data is as follows

This method takes advantage of the Bayesian synthesis inversion feature of analyzing the whole period at once.

Observation data network

Colors indicate the percentage of monthly mean observation values that exist for the analysis period (2003-2020) (left: in-situ observation only, right: satellite added).

In this case, satellite observations were integrated into a 5-degree grid, and grid point values (green circles) with a presence rate of 60% or more for the valid period (2000-2020) were introduced into the inverse analysis.

Satellite observation data (equivalent to 382 stations) were successfully introduced for the land and some ocean areas.

Satellite data were available for tropical regions except tropical South America, but satellite data were scarce for the Southern Hemisphere oceans.

�

CNTL experiment

RAW, JMA, FIX, ALL, MAV experiment

CO2 flux analysis (global)

Above figures show 12-month moving averages of carbon dioxide budgets over land (left) and ocean (right).

​

Ocean shows deviation from a priori information (about -2.1 PgC/yr).

​

CNTL and RAW show relatively close CO2 balance variations, and the remaining balance variations are also close.

Regional CO2 flux

The changes are large in areas where satellite observation data account for a large proportion (e.g., southern South America (L04), Africa (L05, L06), etc.), but in other areas, the changes in the CO2 balance due to the introduction of satellite observation data are small compared to previous studies.

Annual mean CO2flux（2010-2019）

RAW has the smallest overall difference from CNTL due to the tendency for low density in the ocean

​

Global uniform correction (FIX) shows the largest difference from CNTL

​

Experiments with spatial distribution of bias correction (JMA, ALL, MAV) yielded relatively close results.

Validation against independent observations (global)

Independently observed data (CONTRAIL; Machida et al., 2008) and model grid point (TL159L60) units were compared (2010-2020).

​

Differences were largest for CNTL, followed by RAW and FIX, with the remaining three experiments (JMA, ALL, and MAV) at about the same level.

​

The differences after satellite bias correction (JMA, ALL, FIX, and MAV) became smaller from 900hPa to 200hPa.

​

Root-mean-square errors were slightly larger with the introduction of satellite data, but showed little difference from CNTL.

Comparison with CONTRAIL

Independent observation data (CONTRAIL; Machida et al., 2008) and model grid point (TL159L60) units were compared (2010-2020).

(a): CO2 concentration bias between in-situ (CNTL) inversion results and CONTRAIL observation data at 250 hPa.

(b)-(f): Indicates the amount of reduction in bias relative to the CNTL experiment. Unit: ppm.

​

The RAW experiment did little to reduce the bias against CONTRAIL compared to CNTL.

​

The other experiments were able to reduce the bias toward independent observations, mainly in the Northern Hemisphere.

​

The reason for this may be that the CO2 fluxes could not be adequately constrained by the observation data due to the relative lack of data from the Southern Hemisphere in current analysis system.

Summary

We conducted a study on regional CO2 budget analysis using bias-corrected satellite observation data (GOSAT SWIR L2 V2.97-8 + bias correction by our own analysis).

​

The introduction of satellite observations increased ocean sink (-2.3-3.0PgC/yr), although the regional CO2 budget fluctuations were small except for a few regions that were less constrained by field observations.

​

After the introduction of bias-corrected satellite observations, the difference with observations became smaller in the 900-200hPa region in the northern hemisphere, but the difference became larger in the southern hemisphere due to lack of observations.

​

Although only a single bias-corrected satellite observations were used in this study, other satellites (e.g. OCO-2, GOSAT-2) can be used in combination to increase the effectiveness of the analysis of satellite observation data.

​

The bias correction method introduced here can be used for other satellites as well (longer observation periods are preferable).

​

The increase in the difference from the observed data in the Southern Hemisphere suggests that the accuracy of the inverse analysis based only on field observations is inadequate.

​

Improvement of inverse analysis is also needed (number of regions, transport models, analysis methods, etc.).

Acknowledgements

The authors thank observational data providers.

​

This research was supported by the JSPS KAKENHI (grant numbers: JP19K12312), Environment Research and Technology Development Fund (JPMEERF21S20810) of the Environmental Restoration and Conservation Agency of Japan from the Ministry of the Environment, Japan.

Reference

Maki, T., K. Kondo, K. Ishijima, T. T. Sekiyama, K. Tsuboi, T. Nakamura, 2023: Independent bias correction method for satellite observation data introduced to CO2 flux inversion. SOLA, 19, 157-164.

Thank you very much!